Pharmaceutical quality traceability decision-making method and system based on dynamic knowledge graph

Abstract

The invention discloses a pharmaceutical quality traceability decision-making method and system based on a dynamic knowledge graph, and relates to the technical field of pharmaceutical quality control. The method comprises the following steps: defining a dynamic knowledge graph containing quality-related entities, wherein the dynamic knowledge graph at least comprises a quality event entity, a production element entity and a time sequence causal relationship between the quality event entity and the production element entity; capturing production data related to quality in real time, and generating event data with a quality rule identifier based on a preset triggering condition; and carrying out dynamic state modeling on the production element entity to generate a state vector containing a time sequence attenuation characteristic and a quality influence weight. By constructing the dynamic knowledge graph and designing the time sequence causal reasoning algorithm, real-time capture, dynamic modeling and causal relationship analysis of production data are achieved, the defects of the traditional technology in real-time performance and relevance are overcome, and an efficient and accurate intelligent solution is provided for pharmaceutical quality traceability.

Description

Technical Field

[0001] The present invention relates to the technical field of pharmaceutical quality control, and particularly to a pharmaceutical quality traceability decision-making method and system based on a dynamic knowledge graph. Background Art

[0002] In the quality control of the pharmaceutical industry, real-time and accurate quality traceability is the core requirement for ensuring drug safety. Traditional technologies rely on static knowledge graphs or isolated databases, and it is difficult to integrate dynamic data in the production process in real time (such as equipment parameter fluctuations, abnormal quality inspection results, batch transfer information, etc.), resulting in significant lag in quality problem tracing. For example, when production equipment parameters are abnormal or quality inspection results deviate, the static system cannot immediately associate multi-source data and locate the root cause. It often requires manual cross-system investigation, which is time-consuming and laborious and prone to missing key associated factors, seriously affecting the response efficiency of quality problems and the accuracy of decision-making. Summary of the Invention

[0003] The purpose of the present invention is to provide a pharmaceutical quality traceability decision-making method and system based on a dynamic knowledge graph. By constructing a dynamic knowledge graph and designing a time-series causal reasoning algorithm, it realizes real-time capture, dynamic modeling, and causal relationship analysis of production data, and solves the deficiencies of traditional technologies in real-time and relevance.

[0004] To achieve the above purpose, the present invention provides the following technical solution: A pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph, including the following steps: Define a dynamic knowledge graph including quality-related entities, where the dynamic knowledge graph at least includes quality event entities, production factor entities, and the time-series causal relationship between the two; Real-time capture production data related to quality, and generate event data with quality rule identifiers based on preset trigger conditions; Perform dynamic state modeling on the production factor entities to generate state vectors including time-series decay characteristics and quality impact weights; Perform two-stage causal consistency verification on the association relationship between the quality event entities and the production factor entities to generate an effective causal relationship chain; Based on the effective causal relationship chain and a preset quality rule library, generate quality traceability decision-making suggestions including compliance basis.

[0005] Preferably, the step of defining a dynamic knowledge graph including quality-related entities, where the dynamic knowledge graph at least includes quality event entities, production factor entities, and the time-series causal relationship between the two, includes: The quality event entity includes an event unique identifier, an associated batch number, a description of non-conforming items, a timestamp of occurrence, and a severity level classification; The production factor entities at least include raw material batch entities, equipment status entities, process step entities, and personnel operation entities. The raw material batch entities include batch numbers, supplier information, and inspection report data. The equipment status entities include equipment numbers, operating parameters, and maintenance records. The process step entities include step numbers, standard values of process parameters, and execution time intervals. The personnel operation entities include operator IDs, operation types, and operation times; The temporal causal relationships include direct causal relationships and indirect causal relationships. The initial confidence level of the direct causal relationships is determined through statistical analysis of historical production data, and the indirect causal relationships are learned through knowledge graph embedding algorithms; The dynamic knowledge graph is stored in a graph database, supporting incremental updates and version management.

[0006] Preferably, the real-time capture of production data related to quality, generating event data with quality rule identifiers based on preset trigger conditions, includes: Real-time collection of process parameters of production equipment through industrial Internet of Things interfaces, obtaining quality inspection data through laboratory information management systems, and obtaining batch information through enterprise resource planning systems; The preset trigger conditions include parameter fluctuation triggers, quality anomaly triggers, and entity change triggers: Parameter fluctuation trigger: Triggered when the absolute value change of the process parameter exceeds a preset static threshold or the relative change rate exceeds a preset dynamic threshold; Quality anomaly trigger: When the quality inspection data contains descriptions of non-conforming items, generating a trigger event through natural language processing algorithms to parse the inspection report; Entity change trigger: Triggered when the batch information conforms to a preset batch number format or the equipment status undergoes a preset type of change; Among them, the event data is represented in a six-tuple structure: ; In the formula, represents the associated entity object, is a timestamp accurate to milliseconds, is the event type, is the standardized parameter value or event description vector, is the unique identifier of the associated quality rule, is the event severity level.

[0007] Preferably, the dynamic state modeling of the production factor entities, generating a state vector including temporal decay characteristics and quality impact weights, includes: Preprocess the historical parameter values of the production factor entities, including outlier detection and repair, missing value filling, and data standardization; Calculate the time-series decay parameter using the exponential moving average algorithm with a forgetting factor. The formula is: ; where is the forgetting factor, which is dynamically calculated using the exponential decay model. The formula is , is the standardized parameter value at time , and the time constant ; In the formula, is the base time constant, is the adjustment coefficient, is the process risk level; Calculate the quality impact weight to reflect the influence degree of the entity state on the quality event. The formula is: ; In the formula, is the sigmoid activation function, is the learnable weight matrix, which is obtained by training with historical quality event data, is the cumulative number of anomalies at time , is the duration of the abnormal state, is the degree of parameter deviation from the standard value; Combine the time-series decay parameter with the quality impact weight

[0008] to generate an entity state vector containing time-series decay characteristics and quality impact weights, which is used to characterize the dynamic change characteristics of the production factor entity in the time series. Preferably, the two-stage causal consistency check includes: The first-stage time logic check includes: Verify that the timestamp of the cause event is earlier than the timestamp of the result event, and the time difference falls within the process-allowed delay window preconfigured according to the process type and risk level; The second-stage modified Granger causality test includes: ; In the formula, To limit the sum of squared residuals of the model, is the sum of squared residuals of the unrestricted model, is the lag order, is the sample size; When exceeds the preset threshold the causal relationship is confirmed, and the threshold is dynamically adjusted according to the data type and sample size, and the calculation formula is: ;

[0009] In the formula, is the basic threshold, is the noise standard deviation, is the signal standard deviation.

[0010] Preferably, based on the effective causal relationship chain and the preset quality rule base, generating a quality traceability decision recommendation including compliance basis, includes: Starting from the quality event entity, using a breadth-first search algorithm with quality weights to search for the causal relationship chain, and its path priority calculation formula is: ;

[0011] In the formula, represents the path priority, is the quality impact weight of the th entity, is the confidence of the th relationship, is the path length; Mapping the abnormal entities in the causal relationship chain to the quality rule base through a preset rule engine; Generating a structured decision report including root cause analysis, handling measures, and preventive suggestions, and the handling measures include equipment-level handling measures, batch-level handling measures, and process-level handling measures.

[0012] Preferably, the root cause analysis includes at least one direct cause and at least one root cause, and each cause is associated with a corresponding evidence chain and confidence; The equipment-level handling measures include equipment suspension, equipment maintenance, equipment calibration, and equipment performance confirmation, and each measure is associated with a specific standard operating procedure number; The batch-level handling measures include batch isolation, batch rework, batch recall, and batch destruction, and each measure is associated with corresponding regulatory provisions and risk assessment reports; The process-level handling measures include process parameter adjustment, process step optimization, and process verification, and each measure is associated with historical process verification data and statistical process control analysis results.

[0013] The present invention also provides a pharmaceutical quality traceability decision-making system based on a dynamic knowledge graph, including: A graph construction module for defining a dynamic knowledge graph containing quality-related entities, where the dynamic knowledge graph at least includes quality event entities, production factor entities, and the temporal causal relationship between the two; A data collection module for capturing production data related to quality in real time and generating event data with quality rule identifiers based on preset trigger conditions; A causal reasoning module for dynamically modeling the states of the production factor entities and generating state vectors containing temporal decay features and quality impact weights; The causal reasoning module is also used for performing a two-stage causal consistency check on the association relationship between the quality event entities and the production factor entities to generate an effective causal relationship chain; A decision generation module for generating quality traceability decision-making suggestions containing compliance bases based on the effective causal relationship chain and a preset quality rule library.

[0014] The present invention also provides an electronic device, which is an entity device and includes: A processor and a memory, where the memory is communicatively connected to the processor; The memory is used for storing at least one executable instruction executed by the processor, and the processor is used for executing the executable instruction to implement the pharmaceutical quality traceability decision-making method based on the dynamic knowledge graph as described above.

[0015] The present invention also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, it implements the pharmaceutical quality traceability decision-making method based on the dynamic knowledge graph as described above.

[0016] Compared with the prior art, the beneficial effects of the present invention are: Through the dynamic knowledge graph and temporal causal reasoning, the present invention realizes the real-time traceability and compliance decision-making of pharmaceutical quality, can capture production data in real time and update the graph, and quickly respond to quality anomalies; accurately locates the root cause of quality problems through two-stage causal verification, improving the accuracy of traceability; the rule engine generates compliance handling suggestions, reducing the cost of manual review; integrates multi-source production factors to build a full-chain traceability system, supporting the whole-process control from raw materials to finished products, providing an efficient and reliable quality control tool for pharmaceutical enterprises, and enhancing the quality safety guarantee ability and compliance level. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a main flowchart of a pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a pharmaceutical quality traceability decision-making system based on a dynamic knowledge graph provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. Specific embodiments

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] The execution subject of the method in this embodiment is a terminal, and the terminal can be a device such as a mobile phone, a tablet computer, a personal digital assistant (PDA) of a palm computer, a notebook or a desktop computer. Of course, it can also be other devices with similar functions, and this embodiment is not limited.

[0020] Please refer to Figure 1 , the present invention provides a pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph, including: Step 100, define a dynamic knowledge graph including quality-related entities, and the dynamic knowledge graph at least includes quality event entities, production factor entities, and the temporal causal relationship between the two.

[0021] Among them, the defining of the dynamic knowledge graph including quality-related entities, and the dynamic knowledge graph at least includes quality event entities, production factor entities, and the temporal causal relationship between the two, includes: The quality event entity includes an event unique identifier, an associated batch number, a description of non-conforming items, a timestamp of occurrence, and a severity level classification; The production factor entity at least includes a raw material batch entity, an equipment status entity, a process step entity, and a personnel operation entity. The raw material batch entity includes a batch number, supplier information, and inspection report data. The equipment status entity includes an equipment number, operating parameters, and maintenance records. The process step entity includes a step number, standard values of process parameters, and an execution time interval. The personnel operation entity includes an operator ID, an operation type, and an operation time; The temporal causal relationship includes a direct causal relationship and an indirect causal relationship. The initial confidence level of the direct causal relationship is determined through statistical analysis of historical production data, and the indirect causal relationship is learned through a knowledge graph embedding algorithm; The dynamic knowledge graph is stored using a graph database, supporting incremental updates and version management.

[0022] Specifically, define a five - tuple graph structure that includes the core entity of "quality event". .

[0023] Among them, is a set of entities, covering key elements in the entire production process: Raw material batch: As the basic unit for quality traceability, its attributes include batch number (unique identifier), supplier (associated with qualification certification information), and test results (quantitative indicators such as microbial content and purity data); Equipment status entity: Reflects real - time production conditions, including equipment ID (unique equipment code), temperature / pressure parameters (dynamically monitored data), and running time (cumulative working hours, used to predict maintenance needs); Quality event: As the core tracking object, its attributes include event ID (automatically generated unique code), batch number (associated with specific production batches), non - conforming items (such as excessive particle size, unqualified content, etc.), and occurrence time (time stamp accurate to the second level); Supplementary entities also include operators (recording work numbers and job permissions), process steps (including step numbers and operation specification document links), etc.; is a set of relationships, constructing a production logic network: Causal relationship: Constructed based on historical data mining and expert knowledge. For example, "continuous high temperature in the reactor → decrease in product purity", "filter blockage → excessive impurity residue", and the causal probability is quantified through algorithms such as Bayesian networks; Temporal relationship: Follows the production process flow. For example, "raw material mixing → heating reaction → cooling crystallization", and the process connection is visually presented through a Gantt chart to ensure the compliance of the production sequence; And optionally add subordinate relationships (such as "equipment ID → production line number"), collaborative relationships (such as "operator A and operator B jointly complete a certain operation") to improve the relationship network; is a set of attributes, used to support precise quality control: Time stamp: Adopts the ISO 8601 standard format to record the creation, update time of entities and the occurrence time of relationships, supporting nanosecond - level time precision; Quality standard threshold: In addition to basic parameters such as pH value, it is extended to the threshold of critical quality attributes (CQA), such as the tablet hardness range and the injection osmotic pressure standard, and combines real - time monitoring data to trigger an early warning mechanism; Supplementary metadata attributes such as measurement units (such as ℃, MPa) and data sources (sensor numbers or manual entry identifiers) are added to enhance data credibility; is the time dimension, used to record the effective time of entities and relationships: Adopt time window technology, slice and analyze according to dimensions such as hours, shifts, batches, etc. For example, count the correlation frequency between equipment failures and quality events for each shift; Support historical version backtracking, record the change history of the graph structure and data through timestamps, and meet the requirements of audit tracking; Integrate predictive time series, predict future quality trends based on historical data, such as predicting the failure probability next week according to equipment operation parameters; For the quality rule set, realize intelligent association of regulations: Directly associate with the original text of regulations such as "GMP" and "ICH Q7". For example, "microbial over - standard in quality events" automatically matches the requirements for microbial limits in the clean area in the appendix of "GMP"; Build a rule engine to support custom business rules, such as "when the test results of a raw material batch from a certain supplier are close to the threshold for 3 consecutive times, trigger an enhanced inspection process"; Realize dynamic update and expansion of rules through semantic networks to ensure that the system continuously complies with the latest regulatory requirements.

[0024] Furthermore, the temporal - causal relationship covers two types: direct causal relationship and indirect causal relationship.

[0025] Among them, the mining of direct causal relationship adopts statistical analysis methods based on historical production data. By calculating means such as Pearson correlation coefficient and Granger causality test, the correlation between parameters of each link in the production process and the final drug quality index is quantitatively evaluated to determine the initial confidence level; The acquisition of indirect causal relationship depends on the knowledge graph embedding algorithm. This algorithm maps structured data such as production process knowledge, equipment parameter knowledge, and quality standard knowledge into a low - dimensional vector space, uses models such as TransE and ComplEx to learn the potential semantic associations between entities, and then discovers the indirect causal paths hidden in the complex production network.

[0026] Furthermore, the dynamic knowledge graph is constructed based on the Neo4j graph database and adopts a versioned data management scheme: Incremental update: Real - time capture data changes in the production system through CDC (Change Data Capture) technology, and automatically update graph nodes and relationships. Version management: Generate graph snapshots for each major process change and equipment transformation, support multi - version comparison and analysis, and retain the knowledge evolution track. Query optimization: Build composite indexes and path indexes to achieve sub - second query responses for millions of entity relationships, and meet the requirements of real - time decision - making.

[0027] Step 200: Real-time capture of production data related to quality, and generation of event data with quality rule identifiers based on preset trigger conditions.

[0028] Specifically, Step 200 includes: Real-time collection of process parameters such as those of production equipment through industrial Internet of Things interfaces, obtaining quality inspection data through laboratory information management systems, and obtaining batch information through enterprise resource planning systems; The preset trigger conditions include parameter fluctuation trigger, quality anomaly trigger, and entity change trigger: Parameter fluctuation trigger: Triggered when the absolute value change of the process parameter exceeds a preset static threshold or the relative change rate exceeds a preset dynamic threshold; Quality anomaly trigger: When the quality inspection data contains descriptions of non-conforming items, generate a trigger event by parsing the inspection report through natural language processing algorithms; Entity change trigger: Triggered when the batch information conforms to a preset batch number format or when a preset type of change occurs in the equipment status; Among them, the event data is represented in a six-tuple structure: ;

[0029] In the formula, represents an associated entity object, such as a specific production equipment, inspection sample, or drug batch, is a timestamp accurate to milliseconds, ensuring high resolution of the event time series, is the event type, including "parameter fluctuation", "quality anomaly", "entity change", etc., is the standardized parameter value or event description vector. The parameter values are uniformly converted to standard units, and the event descriptions are text vectors encoded in UTF-8, is the unique identifier of the associated quality rule, linked to the specific provisions in the enterprise quality management specification, is the event severity level, divided into levels 1-5 (level 1 is minor, level 5 is severe), and is automatically calculated by a preset risk assessment model.

[0030] It is understandable that in the pharmaceutical production process, process parameters such as temperature, pressure, and rotation speed of production equipment are real-time collected through industrial Internet of Things (IIoT) interfaces. These parameters directly affect the quality and efficiency of drug production. Quality inspection data, including key inspection results such as drug ingredient content and microbial indicators, is obtained through laboratory information management systems (LIMS). At the same time, batch information, covering drug production batch numbers, production dates, production teams, etc., is obtained through enterprise resource planning systems (ERP).

[0031] Among them, the preset trigger conditions include parameter fluctuation trigger, quality anomaly trigger, and entity change trigger: Parameter fluctuation trigger: To accurately monitor the production process, it is triggered when the absolute value change of the process parameter exceeds the preset static threshold or the relative change rate exceeds the preset dynamic threshold. For example, if the temperature in the reaction kettle rises by more than 5°C (static threshold) within 10 minutes, or the temperature rise rate per minute exceeds 1°C (dynamic threshold), the system will automatically trigger an event record; Quality anomaly trigger: When the quality inspection data contains descriptions of unqualified items, a trigger event is generated by parsing the inspection report through natural language processing (NLP) algorithms. The system uses named entity recognition (NER) and rule matching technologies to extract keywords of unqualified items, such as "microbial over - standard" and "content not meeting the standard", from the inspection report text and automatically generates a quality anomaly event; Entity change trigger: It is triggered when the batch information conforms to the preset batch number format (such as the standard format change of YYYYMMDD - XXX) or the device status undergoes a preset type of change (such as the device switches from the "running" state to the "fault" state). The device status change information is collected in real - time through the OPC UA protocol to ensure that change events are captured by the system in a timely manner.

[0032] Step 300: Dynamically model the state of the production factor entity to generate a state vector containing time - series decay characteristics and quality impact weights.

[0033] Specifically, step 300 includes: Step 310: In the pharmaceutical production process, there are often data quality problems with the historical parameter values of the production factor entity. First, pre - process the historical parameter values of the production factor entity, including outlier detection and repair, missing value filling, and data standardization; Outlier detection and repair: Adopt the 3σ principle based on statistics and the isolation forest algorithm to identify and correct abnormal data points that deviate from the normal range to ensure data accuracy; Missing value filling: For parameters with missing values, according to the data distribution characteristics, use mean filling for data conforming to the normal distribution, and use the K - nearest neighbor algorithm (KNN) for non - normal distribution data to fill in missing values based on the characteristics of similar samples; Data standardization: Use the Z - score standardization method to uniformly map parameter values with different dimensions to a standard distribution space with a mean of 0 and a standard deviation of 1, eliminating the impact of dimension differences on subsequent calculations; Step 320: After completing the data pre - processing, in order to effectively capture the dynamic change characteristics of the production factor entity parameters over time, use the exponential smoothing algorithm with a forgetting factor to calculate the time - series decay parameter. The formula is: ;

[0034] wherein is the forgetting factor, which is dynamically calculated using an exponential decay model, and the formula is , is the normalized parameter value at time The time constant ;

[0035] In the formula, is the base time constant, which is an initial value preset according to the general characteristics of the process, is the adjustment coefficient, which is used to control the influence degree of the risk level on the time constant, is the process risk level, which is comprehensively evaluated by quality experts in pharmaceutical enterprises in combination with historical quality data, process complexity and other factors. The value range is usually [0, 1], and the larger the value, the higher the risk; In order to quantify the influence degree of the entity state of each production factor on the quality event, calculate the quality influence weight , and the calculation formula is: ;

[0036] In the formula, is the sigmoid activation function, which maps the calculation result to the interval (0, 1) to facilitate intuitively representing the influence degree, is the learnable weight matrix, which is obtained by collecting a large amount of historical quality event data and training and optimizing using the backpropagation algorithm, so that the weight matrix can accurately reflect the contribution degree of each factor to the quality event; is the cumulative number of anomalies at time is the duration of the abnormal state, is the degree of deviation of the parameter from the standard value. The number of anomalies refers to the number of times that the production factor parameters exceed the normal range within the statistical period; the duration is the duration of the parameter abnormal state; the parameter deviation degree represents the deviation ratio between the actual value and the standard value of the parameter. These three indicators characterize the abnormal situation of the production factor entity and its potential impact on quality from different dimensions.

[0037] Optionally, the time series decay parameter and the quality influence weight Combine to generate an entity state vector containing temporal decay features and quality impact weights, specifically through the formula Calculate, where is the generated entity state vector. By coupling the decay features in the temporal dimension with the impact weights in the quality dimension, this vector can fully characterize the dynamic change features of production factor entities in the time series. It not only reflects the importance decay trend of each production factor over time but also superimposes the weight impact of quality factors on the entity state, thereby providing multi-dimensional feature inputs for subsequent production process analysis, resource optimization allocation, and abnormal state monitoring.

[0038] Step 400: Perform a two-stage causal consistency check on the association relationship between the quality event entity and the production factor entity to generate an effective causal relationship chain.

[0039] Specifically, the two-stage causal consistency check includes a first-stage time logic check and a second-stage modified Granger causality test.

[0040] Among them, the first-stage time logic check includes: Verify that the timestamp of the cause event is earlier than the timestamp of the result event , and the time difference falls within the delay window allowed by the process . The delay window is pre-configured according to the process type and risk level.

[0041] It can be understood that in the quality traceability of the pharmaceutical process, the time logic check is an important basic step in determining causal relationships. This step aims to verify the temporal sequence and the rationality of the time interval between the cause event and the result event. Specifically, it is necessary to strictly verify that the timestamp of the cause event must be earlier than the timestamp of the result event, which conforms to the basic time logic of causal relationships - the cause must precede the result. In addition, the time difference between the two needs to be within the delay window pre-configured by the process. This delay window is not fixed but is dynamically set according to different process types (such as fermentation processes, synthesis processes, etc.) and risk levels. For example, for the aseptic preparation production process with a higher risk level, the delay window setting will be more stringent to ensure that the causal relationships between each link can be accurately captured and to avoid deviations in quality problem tracing caused by incorrect time judgments.

[0042] Among them, in the second-stage modified Granger causality test, the modified Granger causality test calculates the causal statistic through a specific formula to quantitatively judge whether there is a causal relationship between two events, including: Calculate the causal statistic: ;

[0043] In the formula, is the restricted model residual sum of squares, which is the total sum of squared errors generated by the model fitting the data under the constraint that it is assumed that a certain variable has no causal effect on another variable. This value reflects the degree of the model's interpretation of the data under the assumption of excluding a specific causal relationship. It is the total sum of squared errors generated by the model fitting the data under the constraint that it is assumed that a certain variable has no causal effect on another variable, and this value reflects the degree of the model's interpretation of the data under the assumption of excluding a specific causal relationship; is the unrestricted model residual sum of squares, that is, the total sum of squared errors of the model fitting the data without imposing any causal relationship constraints. Compared with the restricted model, it can capture the data characteristics more freely, and usually the value will be less than the restricted model residual sum of squares; is the lag order, which is used to consider the influence duration or influence period of historical data on the current event. In the scenario of pharmaceutical quality traceability, the determination of the lag order needs to combine the periodic characteristics of the specific process flow. For example, in some chemical reaction processes, it may be necessary to consider the influence of the raw material quality data of the previous several batches on the quality of the current batch; is the sample size, which represents the number of data samples used for analysis. The size of the sample size directly affects the reliability of the statistical results. In practical applications, it is necessary to ensure that the sample size is sufficient to meet the validity requirements of statistical tests; When exceeds the preset threshold the causal relationship is confirmed. The threshold is dynamically adjusted according to the data type and sample size, and the calculation formula is: ; In the formula, is the base threshold, which is the initial threshold preset according to industry standards, process experience or historical data, and provides a reference benchmark for causal relationship judgment; is the noise standard deviation, which is used to measure the size of random fluctuations or interference factors in the data. In pharmaceutical production data, noise may come from measurement errors of equipment sensors, small fluctuations in environmental factors, etc.; is the signal standard deviation, which represents the degree of fluctuation of the true and effective information in the data. By adjusting the base threshold through the ratio of the two, the data characteristics under different production environments can be adapted, making the causal relationship judgment more accurate and reliable.

[0044] Step 500, based on the effective causal relationship chain and the preset quality rule library, generate quality traceability decision suggestions including compliance bases.

[0045] Specifically, step 500 includes: Step 510: Starting from the quality event entity, search for the causal relationship chain using a breadth-first search algorithm with quality weights. The path priority calculation formula is as follows: ; In the formula, represents the path priority, is the quality impact weight of the th entity. This weight is obtained by evaluating the importance of critical quality attributes and critical process parameters in the drug production process, and its value range is 0.1 - 1.0. The higher the value, the greater the impact on drug quality; is the confidence level of the th relationship, which is dynamically calculated through a Bayesian network or machine learning model trained with historical production data. The value range is 0.0 - 1.0, and it is used to quantify the reliability of the causal relationship. is the path length. The path length is measured by the number of edges in graph theory to ensure that the path complexity is incorporated into the priority evaluation; Step 520: Map the abnormal entities in the causal relationship chain to the quality rule library through a preset rule engine. The rule library is structured and stored using a three-level classification system: Regulatory layer: It includes the whole-life cycle management clauses of Good Manufacturing Practice for Drugs (GMP) and the quality requirements in the general chapters and monographs of the Chinese Pharmacopoeia; Enterprise layer: It integrates compliance documents such as the enterprise's internal SOP files, process validation reports, and deviation handling records; Risk layer: It integrates the evaluation results of FMEA (Failure Mode and Effects Analysis), constructs a risk-rule mapping matrix, and the mapping process combines forward reasoning and backward verification to ensure that each abnormal entity can obtain at least 3 cross-verified regulatory bases; Step 530: Generate a structured decision report including root cause analysis, handling measures, and preventive suggestions. The handling measures include equipment-level handling measures, batch-level handling measures, and process-level handling measures; Root cause analysis: Use the 5Why analysis method to gradually break down the causal relationship chain and visually present the weight distribution of each influencing factor using a fishbone diagram; The handling measures include: Equipment-level: It includes calibration cycle adjustment, sensor accuracy upgrade, suggestions for replacing key components, etc.; Batch-level: It covers risk control plans such as quarantine release, sampling and re-inspection, and scrapping disposal; Process-level: It involves process improvement strategies such as optimizing parameter thresholds, adjusting process sequences, and adding new intermediate control points; Preventive suggestions: Based on Fault Tree Analysis (FTA), propose a deployment plan for the intelligent monitoring system, establish a dynamic adjustment mechanism for the quality risk warning threshold, and also formulate a supporting training plan for employees' skill improvement.

[0046] Optionally, the root cause analysis includes at least one direct cause and at least one root cause, and each cause is associated with a corresponding evidence chain and confidence level; The equipment-level handling measures include equipment suspension, equipment maintenance, equipment calibration, and equipment performance confirmation, and each measure is associated with a specific standard operating procedure number; The batch-level handling measures include batch isolation, batch rework, batch recall, and batch destruction, and each measure is associated with the corresponding regulatory provisions and risk assessment reports; The process-level handling measures include process parameter adjustment, process step optimization, and process verification, and each measure is associated with historical process verification data and statistical process control analysis results; The preventive suggestions include short-term preventive measures and long-term preventive measures. The short-term preventive measures are implemented within 24 hours, and the long-term preventive measures are discussed in the quality review meeting and incorporated into the process specification revision plan.

[0047] Specifically, the root cause analysis adopts a hierarchical decomposition mechanism to gradually deconstruct the quality abnormal event into at least one direct cause and at least one root cause. A complete evidence chain is constructed for each cause node, covering multi-source data such as production logs, sensor data, and inspection reports, and a confidence level value of 0% - 100% is calculated and marked through the Bayesian network algorithm. For example, when it is found that the drug impurities exceed the standard, the direct cause may be the temperature fluctuation of the reaction kettle, and the root cause may be that the PID parameters of the temperature control system are not updated in time. The corresponding evidence chain includes the real-time temperature curve and calibration records, and the confidence level is 92%. The equipment-level handling measures establish a standardized execution system: Equipment suspension: The triggering conditions include critical parameter out-of-tolerance, safety interlock alarm, etc.; Equipment maintenance: Implement according to the preventive maintenance plan; Equipment calibration: Follow the ISO 17025 standard; Equipment performance confirmation: Adopt the OQ / PQ verification plan to generate a verification report; The batch-level handling measures implement a hierarchical response mechanism: Batch isolation: According to the requirements of the Good Manufacturing Practice for Drugs, set up an electronic fence in the isolation area and associate it with the risk assessment report; Batch rework: The rework plan needs to be approved by the quality department and attached with risk assessment records; Batch recall: Implement the three-level recall procedure of the Measures for the Administration of Drug Recalls, and submit a recall impact analysis report synchronously; Batch destruction: Implement harmless treatment in accordance with the "Regulations on the Safety Management of Hazardous Chemicals". The destruction process shall be videotaped and archived, and associated with the final evaluation. The process-level treatment measures construct a data-driven optimization model: Process parameter adjustment: Based on the DOE experimental design method, referring to the process verification data of the past three years, and combining with the CPK value of the SPC control chart for parameter optimization; Process step optimization: Identify non-value-added steps through value stream mapping (VSM), and associate with the historical process optimization case library; Process verification: Adopt a life cycle verification strategy. The verification plan shall integrate the trend analysis report of historical verification data and the latest SPC process capability analysis results; The prevention suggestions implement a two-track control: Short-term preventive measures: Define a 24-hour golden response period, including immediately increasing the process sampling frequency, starting standby equipment, etc. The implementation of the measures shall be included in the quality exception quick response kanban management; Long-term preventive measures: Include them in the monthly quality review meeting agenda, identify potential risks through FMEA failure mode analysis, and the revised content shall follow the ECN (Engineering Change Notice) process, and synchronously update the process specification document number version.

[0048] In this embodiment, the present invention realizes real-time traceability and compliance decision-making of pharmaceutical quality through a dynamic knowledge graph and temporal causal reasoning, can capture production data in real time and update the graph, and quickly respond to quality exceptions; accurately locate the root cause of quality problems through two-stage causal verification, improving the accuracy of traceability; the rule engine generates compliance handling suggestions, reducing the cost of manual review; integrating multi-source production elements to build a full-chain traceability system, supporting the whole-process control from raw materials to finished products, providing an efficient and reliable quality control tool for pharmaceutical enterprises, and enhancing the quality safety guarantee ability and compliance level.

[0049] Based on the above embodiments, as Figure 2 shown, the present invention also provides a pharmaceutical quality traceability decision-making system based on a dynamic knowledge graph for supporting the pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph in the above embodiments. The pharmaceutical quality traceability decision-making system based on a dynamic knowledge graph includes: A graph construction module 11 for defining a dynamic knowledge graph including quality-related entities, where the dynamic knowledge graph at least includes quality event entities, production element entities, and the temporal causal relationship between the two; A data collection module 12 for capturing production data related to quality in real time and generating event data with quality rule identifiers based on preset trigger conditions; The causal inference module 13 is used to perform dynamic state modeling on the production factor entities, and generate a state vector containing time-series decay features and quality impact weights; The causal inference module 13 is also used to perform two-stage causal consistency verification on the association relationship between the quality event entity and the production factor entity, and generate an effective causal relationship chain; The decision-making generation module 14 is used to generate a quality traceability decision recommendation containing compliance basis based on the effective causal relationship chain and a preset quality rule library.

[0050] In an optional embodiment, the data acquisition module 12 is further used for: Real-time collecting process parameters such as those of production equipment through an industrial Internet of Things interface, obtaining quality inspection data through a laboratory information management system, and obtaining batch information through an enterprise resource planning system.

[0051] In an optional embodiment, the causal inference module 13 is further used for: Preprocessing the historical parameter values of the production factor entities, including outlier detection and repair, missing value filling, and data standardization; Calculating the time-series decay parameter using an exponential moving average algorithm with a forgetting factor; Calculating the quality impact weight to reflect the impact degree of the entity state on the quality event; Combining the time-series decay parameter and the quality impact weight to generate an entity state vector containing time-series decay features and quality impact weights, which is used to characterize the dynamic change features of the production factor entities in the time series.

[0052] In an optional embodiment, the causal inference module 13 is also used for: The first-stage time logic verification, including: verifying that the timestamp of the cause event is earlier than the timestamp of the result event, and the time difference falls within the delay window allowed by the process, and the delay window is pre-configured according to the process type and risk level; The second-stage modified Granger causality test, including: calculating the causal statistic; when the causal statistic exceeds a preset threshold, confirming the causal relationship, and the threshold Is dynamically adjusted according to the data type and sample size.

[0053] In an optional embodiment, the decision-making generation module 14 is further used for: Starting from the quality event entity, searching for the causal relationship chain using a breadth-first search algorithm with quality weights; mapping the abnormal entities in the causal relationship chain to the quality rule library through a preset rule engine; generating a structured decision report containing root cause analysis, handling measures, and preventive suggestions, and the handling measures include equipment-level handling measures, batch-level handling measures, and process-level handling measures.

[0054] In an alternative embodiment, the system further includes a visualization interaction module and a system management module; The visualization interaction module is used to display the knowledge graph, the traceability path, and decision-making suggestions, receive user feedback, and update system parameters; The system management module is used to manage user permissions, maintain a quality rule library, configure system parameters, and record system logs.

[0055] In this embodiment, the present invention realizes real-time traceability of pharmaceutical quality and compliance decision-making through a dynamic knowledge graph and temporal causal reasoning, can capture production data in real time and update the graph, and quickly respond to quality anomalies; accurately locates the root cause of quality problems through two-stage causal verification, improving the accuracy of traceability; the rule engine generates compliance handling suggestions, reducing the cost of manual review; integrates multi-source production elements to build a full-chain traceability system, supporting the whole-process control from raw materials to finished products, providing an efficient and reliable quality control tool for pharmaceutical enterprises, and enhancing the quality safety guarantee ability and compliance level.

[0056] Furthermore, the pharmaceutical quality traceability and decision-making system based on a dynamic knowledge graph can run the above-mentioned pharmaceutical quality traceability and decision-making method based on a dynamic knowledge graph. For specific implementation, reference can be made to the method embodiment, which will not be elaborated here.

[0057] Based on the above embodiments, as Figure 3 shown, the present invention also provides an electronic device, which includes: At least one processor 22, at least one memory 21, a communication interface 23, and a communication bus 24, and the processor 22 is communicatively connected to the memory 21; In this embodiment, the memory 21 can be implemented in any suitable manner. For example: the memory 21 can be a read-only memory, a mechanical hard disk, a solid-state drive, or a USB flash drive, etc.; the memory 21 is used to store executable instructions executed by at least one of the processors; In this embodiment, the processor 22 can be implemented in any suitable manner. For example, the processor 22 can take the form of, for example, a microprocessor or a processor, and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc.; the processor is used to execute the executable instructions to implement the above-mentioned pharmaceutical quality traceability and decision-making method based on a dynamic knowledge graph.

[0058] Based on the above embodiments, the present invention further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method for pharmaceutical quality traceability decision-making based on a dynamic knowledge graph as described above is implemented.

[0059] Those of ordinary skill in the art can realize that the modules and method steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0060] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described devices, equipment, and modules can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0061] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation. For example, multiple modules or units can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or equipment can be in an electrical, mechanical, or other form.

[0062] The modules described as separate components may or may not be physically separated. The components displayed as modules may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0063] In addition, the functional modules in each embodiment of the present invention can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0064] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only storage server, random access storage server, disk or optical disk, and other media that can store program instructions.

[0065] In addition, it should be noted that the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0066] It should be noted that the above examples are only specific embodiments of the present invention, and the present invention is obviously not limited to the above examples, and there are many similar variations. All variations directly derived or associated from the contents disclosed by the technicians in this field should fall within the protection scope of the present invention.

[0067] The above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph, characterized in that Including the following steps: Define a dynamic knowledge graph containing quality-related entities, where the dynamic knowledge graph at least includes quality event entities, production factor entities, and the temporal causal relationship between the two; Real-time capture production data related to quality, and generate event data with quality rule identifiers based on preset trigger conditions; Perform dynamic state modeling on the production factor entities to generate a state vector containing temporal decay characteristics and quality impact weights; Perform a two-stage causal consistency check on the association relationship between the quality event entities and the production factor entities to generate an effective causal relationship chain; Based on the effective causal relationship chain and a preset quality rule library, generate quality traceability decision suggestions containing compliance bases.

2. The pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to claim 1, wherein, The step of defining a dynamic knowledge graph containing quality-related entities, where the dynamic knowledge graph at least includes quality event entities, production factor entities, and the temporal causal relationship between the two, includes: The quality event entity includes an event unique identifier, an associated batch number, a description of nonconforming items, an occurrence timestamp, and a severity level classification; The production factor entities at least include raw material batch entities, equipment status entities, process step entities, and personnel operation entities. The raw material batch entity includes a batch number, supplier information, and inspection report data. The equipment status entity includes an equipment number, operating parameters, and maintenance records. The process step entity includes a step number, standard values of process parameters, and an execution time interval. The personnel operation entity includes an operator ID, an operation type, and an operation time; The temporal causal relationship includes direct causal relationship and indirect causal relationship. The initial confidence level of the direct causal relationship is determined through statistical analysis of historical production data, and the indirect causal relationship is learned through a knowledge graph embedding algorithm; The dynamic knowledge graph is stored using a graph database, supporting incremental updates and version management.

3. The pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to claim 1, characterized in that, The step of real-time capture production data related to quality, and generate event data with quality rule identifiers based on preset trigger conditions, includes: Real-time collect process parameters of production equipment through an industrial Internet of Things interface, obtain quality inspection data through a laboratory information management system, and obtain batch information through an enterprise resource planning system; The preset trigger conditions include parameter fluctuation trigger, quality anomaly trigger, and entity change trigger: Parameter fluctuation trigger: Triggered when the absolute value change of the process parameter exceeds a preset static threshold or the relative change rate exceeds a preset dynamic threshold; Quality anomaly trigger: When the quality inspection data contains a description of nonconforming items, a trigger event is generated by parsing the inspection report through a natural language processing algorithm; Entity change trigger: Triggered when the batch information conforms to a preset batch number format or the equipment status undergoes a preset type of change; Among them, the event data is represented in a six-tuple structure: ； In the formula, represents the associated entity object, is a timestamp accurate to the millisecond level, is the event type, is the standardized parameter value or event description vector, is the unique identifier of the associated quality rule, is the event severity level.

4. The pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to claim 1, wherein, The step of performing dynamic state modeling on the production factor entities to generate a state vector containing temporal decay characteristics and quality impact weights, includes: Preprocess the historical parameter values of the production factor entities, including outlier detection and repair, missing value filling, and data standardization; Use an exponential smoothing algorithm with a forgetting factor to calculate the temporal decay parameter, and the formula is: ； where is the forgetting factor, which is dynamically calculated using an exponential decay model, and the formula is , is the normalized parameter value at time which is dynamically adjusted according to the process risk level, and the calculation formula is: ； In the formula, is the basic time constant, is the adjustment coefficient, is the process risk level; The calculated quality impact weight reflects the influence degree of the entity state on the quality event, and the calculation formula is as follows: ； In the formula, is the sigmoid activation function, is the learnable weight matrix, which is obtained by training with historical quality event data, is the cumulative number of anomalies at time is the duration of the abnormal state, is the degree of parameter deviation from the standard value; Combine the timing decay parameter with the quality impact weight to generate an entity state vector containing the timing decay feature and the quality impact weight, which is used to characterize the dynamic change feature of the production factor entity in the time series.

5. The pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to claim 1, characterized in that, The two-stage causal consistency verification includes: The first-stage time logic verification includes: Verify the timestamp of the cause event is earlier than the timestamp of the result event , and the time difference falls within the delay window allowed by the process , and the delay window is pre-configured according to the process type and risk level; The second-stage modified Granger causality test includes: Calculating the causal statistic: ； Wherein, is the restricted model residual sum of squares, is the unrestricted model residual sum of squares, is the lag order, is the sample size; When exceeding a preset threshold the causal relationship is confirmed, and the threshold is dynamically adjusted according to the data type and the sample size, and the calculation formula is: ； Wherein, is the basic threshold value, is the standard deviation of noise, is the standard deviation of the signal.

6. The pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to claim 1, characterized in that Based on the effective causal relationship chain and the preset quality rule library, generating quality traceability decision suggestions including compliance bases, including: Starting from the quality event entity, using a breadth-first search algorithm with quality weights to search for the causal relationship chain, and the path priority calculation formula is: ； In the formula, represents the path priority, is the quality impact weight of the th entity, is the confidence of the th relationship, is the path length; Mapping the abnormal entities in the causal relationship chain to the quality rule library through a preset rule engine; Generating a structured decision report including root cause analysis, handling measures, and preventive suggestions, where the handling measures include equipment-level handling measures, batch-level handling measures, and process-level handling measures.

7. The pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to claim 6, wherein The root cause analysis includes at least one direct cause and at least one root cause, and each cause is associated with a corresponding evidence chain and confidence level; The equipment-level handling measures include equipment suspension, equipment maintenance, equipment calibration, and equipment performance confirmation, and each measure is associated with a specific standard operation procedure number; The batch-level handling measures include batch isolation, batch rework, batch recall, and batch destruction, and each measure is associated with corresponding regulatory clauses and risk assessment reports; The process-level handling measures include process parameter adjustment, process step optimization, and process verification, and each measure is associated with historical process verification data and statistical process control analysis results.

8. A pharmaceutical quality traceability decision-making system based on a dynamic knowledge graph, characterized in that, Including: A graph construction module for defining a dynamic knowledge graph containing quality-related entities, where the dynamic knowledge graph at least includes a quality event entity, a production factor entity, and the temporal causal relationship between the two; A data collection module for capturing production data related to quality in real time and generating event data with quality rule identifiers based on preset trigger conditions; A causal reasoning module for dynamically modeling the state of the production factor entity and generating a state vector containing temporal decay features and quality impact weights; The causal reasoning module is also used for performing two-stage causal consistency verification on the association relationship between the quality event entity and the production factor entity to generate an effective causal relationship chain; A decision generation module for generating quality traceability decision suggestions including compliance bases based on the effective causal relationship chain and the preset quality rule library.

9. An electronic device, characterized in that, The electronic device includes: A processor and a memory, and the memory is communicatively connected to the processor; The memory is used to store at least one executable instruction executed by the processor, and the processor is used to execute the executable instruction to implement the pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, it implements the pharmaceutical quality traceability decision-making method based on a dynamic knowledge graph according to any one of claims 1 to 7.

Images